Troubleshooting a large erlang system

In this paper, we discuss some experiences from a large, industrial software project using a functional programming language. In particular, we will focus on programming errors.The software studied is the AXD 301 (a multi-service switch from Ericsson AB [1]) control system. It is implemented in a functional language Erlang [2 ]. We will discuss ho this affects programmer productivity.There are now well over 1,000 AXD 301's deployed. Even though a properly handled AXD 301 is quite reliable, there exists a great deal of knowledge about problems that do occur in production code. We will analyze what kinds of programming errors cause these problems, and suggest some methods for preventing and, when that fails, finding the errors. We will also describe some tools that has been specifically developed to aid in debugging.One (perceived) problem with using a interpreted, functional language is execution speed. In practice, we have found that the overhead of running in an emulator is not dramatic, and that it is often more than compensated for by the advantages. The expressiveness of the language and the absence of low-level bugs means that programmers have more time to spend on tuning the code. And since the emulator has good support for tracing, one can perform very advanced profiling, thus making the code intrinsically more effective. We will discuss a profiling tool developed for that purpose.